Unverified50% confidenceSolutionExact time
纵深防御原则要求系统安全性不应依赖于用户的完美操作，每一层工具都应假设上一层可能已被突破并提供独立的安全校验
1
Sources
50%
Confidence
Long-term
Relevance
7/21/2026
First Seen
Sources
Related Claims
UnverifiedPrompt-level guardrails (constraining agent behavior solely through system prompts) are inherently fragile, and a more reliable direction is to enforce controls at the tool-call level.73% similarUnverifiedHITL设计应将风险判断逻辑上移至Agent层，而非依赖工具本身提供安全保障，体现关注点分离与纵深防御原则71% similarUnverified系统的整体可靠性由最薄弱的那个环节决定，消除单点故障是保障关键系统持续可用的首要原则68% similarUnverifiedAI models capable of vulnerability discovery may allow people without deep security knowledge to find and exploit vulnerabilities, lowering the barrier to attacks.68% similarUnverified通用性与专业性之间存在张力，专注单一任务的工具通常能达到更高精度和可控性，而通用系统面临每个环节够用但不够好的风险68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/579925API
curl https://kongchang.com/api/v1/knowledge/claims/579925MCP
get_claim(id=579925)